Problem set 6: Bayesian Games

Note: an (*) means that the question is moderately difficult and (**) means that the
question is challenging. No (*) means the question is standard.

1. Firm A is considering taking over firm B. It does not know firm B’s value. It believes
that this value, when firm B is controlled by its own management, is at least $0 and at
most $100, and assigns equal probability to each of the 101 dollar values in this range.

Firm B will be worth 50% more under firm A’s management than it is under its own
management. Suppose whether or not firm A acquires B is determined according to
the following procedure. Firm A makes a sealed bid (nonnegative numbers) and firm B

chooses the lowest offer to accept (nonnegative numbers). If the bid is at least as high
as this lowest offer, A acquires B and pays its bid. Otherwise no sale takes place. If firm
A buys firm B for a price of y when firm A’s current value is x, then firm A’s payoff is
3
2
x − y and firm B’s payoff is y. If no sale takes place, A’s payoff is zero and B’s payoff
is x. Find a Bayesian Nash equilibrium of this game.

2. In each of the two cases below, we have a game with two players, each of whom has
two types. The types of player 1 are T1 = {a, b}. The types of player 2 are T2 = {c, d}.

(a) Suppose the beliefs of the types are p1 (c | a) = 1/4, p1 (c | b) = 3/4, p2 (a | c) = 1/4,
and p2 (a | d) = 3/4. Show that a common prior exists. (hint: The easiest way is to just
construct a prior that works).

(b) Now suppose the beliefs are p1 (c | a) = 1/2, p1 (c | b) = 1/2, p2 (a | c) = 1/3, and

p2 (a | d) = 2/3. Show that there cannot be a common prior.

3. Consider the following game of incomplete information. The two players are either

1
playing the game G1 which is

b1 b2
a1 3,0 0,1
a2 0,0 3,1
a3 2,0 2,1

or the game G2 which is

b1 b2
a1 3,0 0,1
a2 0,1 3,0
a3 2,2 2,0

(a) Assume that player 2 knows which game they are playing, but player 1 does not.
Instead, player 1 views each game as equally likely. Write this as a game of incomplete
information. Find all pure strategy Bayes–Nash equilibria.

(b) Now suppose that player 2 knows which game they are playing but doesn’t know

whether or not player 1 knows. More specifically, the common prior is that 2 always
knows the game, while player 1 knows the game being played with probability 1/2. If
player 1 does not know the game being played, he views the two games as equally likely.

Write this as a game of incomplete information. Find all pure strategy Bayes–Nash
equilibria.

4. Consider two risk-neutral firms that compete in quantities (Cournot). The aggregate
inverse demand is given by P (Q) = 3 − Q. The constant marginal costs are distributed

2
as follows:
(c1 , c2 ) (1, 1) (2, 2) (1, 2) (2, 1)
1 1 1 1
Pr(c1 , c2 ) 3 3 6 6

Each firm can only observe its own cost. Find a symmetric BNE.

5. Consider two risk-neutral firms that compete in quantities (Cournot). The aggregate
inverse demand is given by P (Q) = 3 − Q. The constant marginal cost of firm 2 is zero,
while the constant marginal cost of firm 1 is either 0 or 1 with equal probability. If the

cost is 0, then there is a 50-50 chance that firm 2 knows this. If the cost is 1, then firm
2 does not know this. Firm 1 does not know whether or not firm 2 knows its (firm 1’s)
cost. Find the BNE of this game.

6. Two agents can jointly provide a public good. The cost of providing the good is 3.
The value of the good to agent i is vi and is known only to agent i. The prior of the
agents is that the valuations are drawn independently with the following distribution:

0, with probability 1/2;
vi =
2, otherwise.

(a) Consider the following game. The agents simultaneously offer contributions where
the contribution of agent i is bi ≥ 0. If b1 + b2 ≥ 3 (so the contributions exceed the cost

of providing the good), then the good is provided and the contributions are paid. In this
case, agent i’s payoff is vi − bi . If instead the contributions fall short, so b1 + b2 < 3, then
the public good is not provided but the contributions are kept. In this situation, agent i’s

payoff is −bi . First, write this as a normal form game of incomplete information. Then
find the unique equilibrium of this game.

(b) Now suppose we change the game as follows. As before, the agents simultaneously

offer contributions. Again, if b1 + b2 ≥ 3, the good is provided and agent i’s payoff is

3
vi − bi . In this case, though, if b1 + b2 < 3, the contributions are refunded, so agent
i’s payoff is 0. Write this as a normal form game of incomplete information. Then find

an efficient equilibrium — that is, an equilibrium in which the public good is always
provided if it is efficient to do so (and is not provided otherwise). (Note: There are many
equilibria, so don’t try to get uniqueness.)

7*. Consider a situation of bilateral trade between a seller and a buyer, both of which

are risk-neutral with quasi-linear preferences. The seller has a single item for sale and
his value for it is uniformly distributed on [0, 1]. The buyer’s value for the object is also
uniformly distributed on [0, 1]. The values for the seller and the buyer are independent

of each other. The buyer knows only his own value and the seller knows only his own
value. Assume the buyer and the seller conduct a double auction. The seller makes an
ask ps and the buyer makes a bid pb . These are made simultaneously (so that when one

party makes a decision it does not observe the other party’s decision). If pb > ps a sale is
made at a price p = 12 (ps + pb ). The seller’s payoff in this case is p − vS and the buyer’s
payoff is vb − p. If ps ≥ pb , then there is no sale, the seller’s payoff is vs and the buyer’s
payoff is zero.

(a) Find a BNE in linear strategies, i.e., one where each player i uses the strategy
pi (vi ) = αi + βi vi , where βi > 0.

(b) Given the equilibrium you computed, is the auction efficient? I.e., is it the case that
whenever vs < vb a sale is made?

8*. In an auction with common values, each bidder’s valuation depends on the signals

of other bidders as well as his own. Suppose there are two bidders, each bidder’s signal
ti is independently drawn from a uniform distribution on [0, 1], and the valuation of each

4
bidder i is given by vi = αti + γtj , where j is the other bidder and α ≥ γ ≥ 0.

(a) Show that in a second -price sealed-bid auction there is a BNE in which each type ti
of each player i bids (α + γ)ti .

(b) Show that in a first-price sealed-bid auction there is a BNE in which each type ti of
each player i bids 21 (α + γ)ti .

9*. This question demonstrates what is known as the “swing voter’s curse”.1 Suppose
that whether candidate 1 or 2 is elected depends on the vote of two citizens. The economy

may be in one of two states, A or B. The citizens agree that candidate 1 is best if the state
is A and candidate 2 is best if the state is B. Each citizen’s preferences are represneted by
the expectation of a vNM utility function that assigns a payoff of 1 if the best candidate

for the state wins (i.e., if he obtains more votes than the other candidate), a payoff of
1
zero if the other candidate wins, and a payoff of 2
if the candidates tie. Citizen 1 is
informed of the state, whereas citizen 2 believes it is A with probability 0.9 and B with

probability 0.1. Each citizen may either vote for candidate 1, vote for candidate 2, or not
vote.

(a) Formulate this situation as a Bayesian game.

(b) Show that the game has exactly two pure Nash equilibria, in one of which citizen 2

does not vote, and in the other of which he votes for candidate 1.

(c) Show that one of the player’s actions in the second of these equilibria is weakly

dominated.

(d) Why is “swing voter’s curse” an appropriate name for the determinant of citizen 2’s
1
Timothy J. Feddersen and Wolfgang Pesendorfer (1996), “The Swing Voter’s Curse”, American
Economic Review 86, 408-424.

5
decision in the second equilibrium?

10**. Recall the jury example from the notes. Under some conditions on z the game has

a symmetric mixed strategy equilibrium in which each type g juror votes for conviction,
and each type i juror votes for conviction with probability 0 < β < 1 and for acquittal
with probability 1 − β. Show that in this BNE,

pX − (1 − q)
β=
q − (1 − p)X

where
π(1 − p)(1 − z) n−1
1
X=[ ]
(1 − π)qz

11**. Recall the toy version of the Myerson-Satterthwaite theorem from the notes.

Prove the claim that when η ≤ 2α there exists an efficient double auction.

12**. Ariel Rubinstein’s “e-mail game” is the first example that started the literature
on contagion/global-games.2 Two players are about to play one of two games:

Ga A B Gb A B
A M, M 1, −L A 0, 0 1, −L
B −L, 1 0, 0 B −L, 1 M, M
1−p p< 12

where

L>M >1

The true game is known only to player 1.

(a) Suppose player 1 cannot communicate his information to player 2. Formulate this
situation as a Bayesian game and find its Nash equilibria.
2
Ariel Rubinstein (1989), “The Electronic Mail Game: Strategic Behavior Under ‘Almost Common
Knowledge’”, American Economic Review 79, 385-391.

6
(b) Suppose next that player 1 can perfectly communicate his information to player 2.
Formulate this situation as a Bayesian game and find its Nash equilibria.

(c) Now suppose that 1 has a noisy channel through which he communicates with 2.
In particular, whenever the game is Gb (and only then), his computer sends a signal to

player 2’s computer. Each computer that receives a signal sends a confirmation to the
other computer. Confirmations are also sent when a computer receives a confirmation.
However, for each computer there is a small probability  > 0 that the signal it sent will

not arrive at its destination. Each player’s computer displays the number of messages
(including confirmations) that that computer has sent. Messages are sent instantenously,
so when the number on each player’s computer screen no longer changes, the communi-

cation phase has ended. Each player then needs to choose an action given the number
displayed on his screen.

To analyze this situation as a Bayesian game we need to define a state space and

specify the players’ information (their types). Let a state be the pair of numbers (Q1 , Q2 )
that appear on the players’ computer screens at the end of the communication phase.
Since player 1’s computer is the first to send a message, Q1 is either equal to Q2 (all

messages sent from 1 to 2 were received but the last confirmation from 2 was lost) or
equal to Q2 + 1 (all but the last message from 1 were received by 2).

Ω = {(Q1 , Q2 ) : Q1 = Q2 or Q1 = Q2 + 1}

Player 1’s information function is given by the partition,

{0, 0}, {{1, 0}, {1, 1}}, {{2, 1}, {2, 2}}, . . .}

Player 2’s information function is given by the partition,

{{0, 0}, {1, 0}}, {{1, 1}, {2, 1}}, . . .}

7
The game played in state (0, 0) is Ga , while in all other states it is Gb .

• Player 1 knows the game in every state.

• Player 2 knows the game in all states but (0, 0) and (1, 0).

• In each of the states, (1, 0) and (1, 1), player 1 knows that the game is Gb , but he
doesn’t know whether 2 knows this.

• In each of the states, (1, 1) and (2, 1), player 2 knows that the game is Gb , but he
doesn’t know whether 1 knows that he (player 2) knows this.

• In each state we will have long statements of the form, ”player i knows that player

j knows that player i knows...” but it will always stop at some finite point.

(c1) What are the players’ prior beliefs about the states? To characterize these beliefs
you will need to specify (as a function of p, q and ) what is pi (0, 0) and what are pi (q+1, q)

and pi (q + 1, q + 1).

(c2) Explain why in state (0, 0) player 1 would choose A in any BNE.

(c3) Consider the type of player 2 who sees a zero on his screen. Show that in any BNE
this type would choose A.

(c4) Suppose you proved that for all (Q1 , Q2 ) with Q1 + Q2 < 2q, both players choose

A in any BNE. Consider player 1 who sees the number q on his screen. Prove that this
type of player 1 would choose A in any BNE.

(c5) Suppose you proved that players 1 and 2 choose A in any equilibrium for all (Q1 , Q2 )
with Q1 + Q2 < 2q + 1. Show that 2 would choose A whenever he sees q on his screen.

8
(c6) What is the implication of (c2)-(c5)? Compare this with (a). What is striking
about this result?